[ACL 2025 Demo] ReasonGraph: Visualisation of Reasoning Methods and Extended Inference Paths

March 9, 2026 Β· View on GitHub

Language

English β€’ δΈ­ζ–‡

Content

πŸš€ News β€’ ✏️ Todo β€’ ✨ Introduction

πŸ‘€ Examples β€’ 🎨 Quick Use β€’ πŸ–₯️ Environment

πŸ“Œ Citation β€’ πŸ”– License

Links

Online Demo β€’ Paper β€’ Discord β€’ Questionnaire

Β 

πŸš€ News

  • [2025.05.23] Accepted by ACL 2025 Demo!
  • [2025.03.24] ⭐ Support visualisation of long outputs from reasoning models!!!
  • [2025.03.18] Support more LLM providers and models. Support two languages, English and Chinese.
  • [2025.03.18] Rate and provide comments in the questionnaire (takes about 30 seconds).
  • [2025.03.17] Try the online demo through Huggingface Page!
  • [2025.03.14] Join our Discord group!
  • [2025.03.07] The paper is available in Arxiv.
  • [2025.02.22] Create the Github page.
Β 
Β 
Β 

✏️ Todo & Help Wanted

Β 
Β 
Β 

✨ Introduction

ReasonGraph is an open-source web platform for visualizing and analyzing reasoning processes of Large Language Models (LLMs).

Functions:

  • Visualisation of Reasoning Paths: Transforms text-format reasoning outputs into flow charts.
  • Long Reasoning: Supports visualisation of extended outputs from reasoning models.
  • Meta Reasoning: Provides built-in capabilities for models to self-select optimal reasoning methods.

Characteristics:

  • Model Support: Integrates with over 50 state-of-the-art models from major LLM providers including Anthropic, OpenAI, Google, Grok, Deepseek, Qwen, and Together.AI.
  • Reasoning Methods: Implements mainstream reasoning approaches including sequential methods and tree-based methods.
  • Modular Framework: Standardized APIs for easy integration of new reasoning methods and models.
  • Beginner-Friendly: Intuitive UI design with visualization updates and simple configuration.
  • Multilingual: Support multi-languages and easy to be extended for more languages.
Β 
Β 
Β 

πŸ‘€ Examples

UI Screenshot:

Visualisation of outputs from reasoning models:

Input: Give me two suggestions for transitioning from a journalist to a book editor? Use sequential and tree-based thinking and refinement loops.
Model: deepseek-reasoner from DeepSeek

Β 
Visualisation of sequential reasoning methods:

Chain of Thoughts (top-left), Self-refine (top-middle), Least-to-most (top-right), Self-consistency (bottom-left):

Β 
Visualisation of tree-based reasoning methods:

Plain text (top), Beam Search (middle), Tree of Thoughts (bottom):

Β 
Β 
Β 

🎨 Quick Use

Try the online demo:

1. Go to the website: https://huggingface.co/spaces/ZongqianLi/ReasonGraph

Β 
Visualize long outputs from reasoning models:

1. Select a reasoning model and enter its API key.

  • API Provider: Deepseek; Model: deepseek-reasoner
  • API Provider: Qwen; Model: qwq-plus

2. Enter the API key for Claude.

3. Click "Long Reasoning".

Β 
Install the package:

1. Set up the environment according to Section πŸ–₯️ Environment below.

2. Go to root directory:

<absolute_path>/ReasonGraph/

3. Input the API key:

If you don't enter the API keys, the interface can still run normally, but you won't be able to use the corresponding models for inference.

<absolute_path>/ReasonGraph/api_keys.json

{
    "anthropic": "<to be filled>",
    "openai": "<to be filled>",
    "google": "<to be filled>",
    "together": "<to be filled>"
}

4. Run the program with a single line of code in the terminal:

python app.py

5. Open your browser and go to the local URL shown in the output.

 * Running on all addresses (X.X.X.X)
 * Running on http://XXX.X.X.X:XXXX
 * Running on http://XX.XXX.XXX.XXX:XXXX
Β 
Β 
Β 

πŸ–₯️ Environment

python==3.11.8
requests==2.31.0
openai==1.63.2
together==1.4.1
flask==3.1.0
google==3.0.0
google-genai==1.2.0
google-generativeai==0.8.4
Β 
Β 
Β 

πŸ“Œ Citation

@inproceedings{li-etal-2025-reasongraph,
    title = "{R}eason{G}raph: Visualization of Reasoning Methods and Extended Inference Paths",
    author = "Li, Zongqian  and
      Shareghi, Ehsan  and
      Collier, Nigel",
    editor = "Mishra, Pushkar  and
      Muresan, Smaranda  and
      Yu, Tao",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-demo.14/",
    doi = "10.18653/v1/2025.acl-demo.14",
    pages = "140--147",
    ISBN = "979-8-89176-253-4",
}
Β 
Β 
Β 

πŸ”– License